Module Overview
[Lecture on 1.3]
Professor John Murphy
j.murphy@ucd.ie
You are expected to attend all lectures - they are not recorded and there will be important information not included in the emails.
"Computer Networks, 5th ed"
Prentice Hall 2010
Andrew S. Tanenbaum and Wetherall
Entirely optional - notes should be sufficient.
Problem sets will be given out
They are not graded or handed up, they're practice
Four in-class tests
Provisional dates:
Test 1: Friday October 4th (11 am) 10% (Week 4)
Test 2: Friday October 25th (11 am) 20% (Week 7)
Test 3: Friday November 15th (11 am) 30% (Week 10)
Final Exam (on-campus with laptop) in the exam period - 40%
We'll get letter grades for each one (I think ?) then he uses the calculation points system for final grades.
Two extreme topologies:
Mesh vs. central
Mesh obviously requires way too many links
Central has drawbacks:
So we'll need something inbetween
Two kinds of link:
You get topologies that could look like this:
Weaker alternatives to strictly non-blocking switch
Non-blocking with Rearrangement:
Non-Blocking with Connection Packing Rule
Some blocking may be acceptable if it's small enough.
Basic analogue crosspoint:
Each crosspoint here can be enabled to link the two wires passing under it
Consider that this switch has 512 possible configurations, the vast majority of which are not very useful. This may suggest that there are too many crosspoints (
Modern crosspoints are electronic transistors
See also: Time Switching
[End of lecture on 1.3]
[Lecture on 1.5]
You can apply this idea recursively, turning the center stage (of
Number of crosspoints:
Noting that
Total:
Optimal value is
For
You'll need to round up or down to an integer usually - generally you should attempt both up and down and see which performs better.
The middle stage has the most switches, so we can eliminate some by replacing that with a 3-stage clos switch, yielding a 5-stage switch.
There is an optimal number of stages for a given value of N
Generally you'll only need 3 or 5 stages, it's only worth implementing more for extremely large switches